ADHD edit benchmark: Four models, one clip
Watch Astra 6.2, Fable 5.2, Sol 6.2 and Opus 5.5 edit the same 22-second Herdr clip, then compare their cuts side by side.
Each model had the same source footage, editing rules and 90-minute limit to make a roughly 20-second vertical short. The video plays all four edits in table order, then puts them side by side with the source audio. The order carries no ranking.
Run stats
| Model | Harness | Edit time | Run time | Cuts | Effects | Sound cues | Speech joins |
|---|---|---|---|---|---|---|---|
| Astra 6.2 | Codex | 11.5 min | 19.40 s | 11 | 36 | 12 | 0 |
| Fable 5.2 | Claude Code | 21 min | 20.00 s | 19 | 102 | 34 | 2 |
| Sol 6.2 | Codex | 19 min after restart | 19.30 s | 12 | 54 | 11 | 0 |
| Opus 5.5 | Claude Code | 23 min | 19.50 s | 17 | 88 | 35 | 3 |
Edit time runs from the first status update to completion. Sol lost an additional 16 minutes before a harness restart. Cuts, effects and sound cues were self-reported. Effects include sound cues, and the models counted effects differently.
What each edit emphasizes
- Astra 6.2
- A print-inspired load meter and a paper receipt for the “80 minutes” line.
- Fable 5.2
- Falling agent tiles, split-screen framing and dense sound and caption cues.
- Sol 6.2
- A scattering tab grid, a route through the app and a large typographic time callout.
- Opus 5.5
- A swarm over the sidebar, a glitch transition and punch-ins on the screen as each claim is made.